Construction of Offensive Play Measurement Items and Shot Prediction Model Applying Machine Learning in Japan Professional Football League
Hirotaka Jo, Hiroki Matsuoka, Kozue Ando and Takahiko Nishijima
[Received May 10, 2021; Accepted October 11, 2021]
The demand for sports analytics is increasing because it contributes to the victory of competitive sports. Although the technology for automatically measuring tracking data has improved, it is meaningful to systematize analysis using ball touch data because the stadiums that can be used are limited. Moreover, in recent years, analysis using machine learning has been increasing, and it is necessary to accumulate research. With this background, this study aimed to construct measurement items and a model for predicting to shoot in soccer offensive play by applying machine learning. Using the Delphi method, 6 items were deleted from the old measurement item group in the previous study, 11 items were newly created, and 45 items were set as the new measurement item group. A decision tree, random forest, and gradient boosting decision tree were applied to the new measurement item group, and several shot prediction models were constructed. The results indicated that the model using 23 items in the gradient boosting decision tree was the best. Furthermore, a comparison between the old and new measurement item groups revealed that the prediction accuracy of the new measurement item group was higher. In conclusion, 45 measurement items of soccer offensive play were constructed, and the shot prediction model using them was constructed by applying machine learning.
Keywords: football, offensive play, measurement items, machine learning, shot prediction
[Football Science Vol.19, 1-21, 2022]